Table 10 The training duration, inference duration, GPU (power), and carbon footprint of training on MUSDB18 of RPCA-DRNN methods, sRNN and E-MRP-CNN
From: RPCA-DRNN technique for monaural singing voice separation
| Â | Training duration (h) | Inference duration (h) | GPU (power) | Carbon footprint of training (kg of carbon dioxide) |
|---|---|---|---|---|
E-MRP-CNN | 200 | - | Two 1080Ti + one 2080Ti + one Titan RTX + one Titan V + one Titan XP (1560 W) | 193 |
RPCA-DRNN | 496 | 1 | One 1080Ti (250 W) | 77 |
RPCA-DRNN | 156 | 0.3 | One 3090 (350 W) | 34 |
RPCA-DRNNl | 28 | 0.3 | One 3090 (350 W) | 6 |
RPCA-DRNNf | 28 | 0.3 | One 3090 (350 W) | 6 |
sRNN | 54 | 0.3 | One 3090 (350 W) | 12 |